MLP-Based Power Analysis Attacks with Two-Point Joint Feature Selection

The effective feature selection and classification of power traces in side-channel attacks has been a hot topic of research in recent years. Traditional side-channel attacks, such as simple power analysis attacks and template attacks, require a large number of power traces to break the key, while machine learning algorithms require fewer power traces to successfully recover the key, thus being able to greatly improve the success rate of the attack. Among them, the MLP network proves effective in data classification. Therefore, based on the MLP network, we investigate and propose a power analysis attack method with the feature selection of two joint correlation coefficients. Via comparing the conventional attack methods, the success rate of this attack method is significantly improved.